This paper presents a new software model designed for distributed sonic signal detection runtime using machine learning algorithms called DeLMA. A new algorithm--Acoustic Data-mining Accelerator (ADA)--is also presented. ADA is a robust yet scalable solution for efficiently processing big sound archives using distributing computing technologies. Together, DeLMA and the ADA algorithm provide a powerful tool currently being used by the Bioacoustics Research Program (BRP) at the Cornell Lab of Ornithology, Cornell University. This paper provides a high level technical overview of the system, and discusses various aspects of the design. Basic runtime performance and project summary are presented. The DeLMA-ADA baseline performance comparing desktop serial configuration to a 64 core distributed HPC system shows as much as a 44 times faster increase in runtime execution. Performance tests using 48 cores on the HPC shows a 9x to 12x efficiency over a 4 core desktop solution. Project summary results for 19 east coast deployments show that the DeLMA-ADA solution has processed over three million channel hours of sound to date.
@article{arxiv.1509.03591,
title = {High Performance Computer Acoustic Data Accelerator: A New System for Exploring Marine Mammal Acoustics for Big Data Applications},
author = {Peter Dugan and John Zollweg and Marian Popescu and Denise Risch and Herve Glotin and Yann LeCun and and Christopher Clark},
journal= {arXiv preprint arXiv:1509.03591},
year = {2015}
}
Comments
Seven pages, submitted at International Conference on Machine Learning 2014, Workshop uLearnBio, unsupervised learning for bioacoustic applications